A question I've been pondering for a while. There's no doubt that they are both good. Both defined their formats in standard, and have found spots in modern once they exited standard. The question now becomes, what circumstances set them apart, and is there a superior model here (both in modern)?
As a general description of what makes both of them good: they both retain card advantage, are only slightly conditional, have flash, work to generate control and field presence, are are relatively affordable (mana), and are in general some of the best topdecks available in modern.
The major differences are the cost to pull off the benefits, the limitation of activation requirements, and the field presence created after the flashback mechanic resolves.
SNAP
Snapcaster is cheaper, under most circumstances, as most of the targets you want to flash back are between 1 and 4 costs, most notably stuff like remand, lighting bolt, path to exile, negate, fatal push, serum visions, opt, etc. The big one it wants to pull back is cryptic command, but more on that later. The idea here is that you need 2 mana to snap, and around 2 mana to hit his flashback abil on average with a max needed being around 4, so in total you need between 4 and 6 mana available, which isn't too bad. He is basically useless in the opening hand (need mana and available grave material), but its not a late game card and can help out in the middle.
In terms of limitation of activation, snap's only real problem is grave material as it gets both instants and sorceries. Barring sideboard interference, this shouldn't be a problem at all. Snap can basically pull back anything, nukes, discard stuff, draw stuff, answers to creatures, answers to enchantment/artifact, etc.
For field presence, snap isn't really that great. a 2/1 is by no means nothing, but its not a reliable clock to put opponent on notice, and it can't really do much more than chump block most creatures. Now, it can be combined with a spell like lightning bolt to take down something bigger, but beyond that it is of intermediate value in terms of field presence. Still obviously better than nothing though, and the instant speed with which he hits the field can't be overstated, making the 2/1 aspect of him a little negated by how fast and timely it acts.
GEARHULK
In terms of cost, a 6 drop is pretty steep. Its a brick in the opening and the middle, and is really only viable once both players have completed development and are fighting for their win conditions to resolve. But, since you don't pay anything for the snap mechanic, you know you have to pay 6 no matter what, which helps with predictability. As discussed above, the average available mana you need for snap is between 4 and 6, so the comparison in that aspect is pretty similar. 4 vs 6 means the difference between middle game usefulness though, so having a snap that can pull back a fatal push vs a gearhulk that can pull back a fatal push is the a different story. But, and this is the big but, 4 drops are the same cost for both, most notably, everyone favorite cryptic command.
As for limitations, snap gets you instants and sorceries, gearhulk gets you instants, which is a pretty notable difference. Snap fetches the 4 drop nukes, the other variations on a nuke, discard outlets, and all the other fun stuff that sorceries give you in modern. Gearhulk is stuck with instants, which is a decent curb, but still not a killshot on his usefulness, as instants in modern are pretty prevalent and you can build around his lack of sorcery grabbing.
As for field presence, this is where the world of difference comes in. 5/6 at instant speed is a game changer by itself. The ability to flash in a big critter and potentially block off a threat, PLUS the card advantage he gives with the snap effect is a huge consideration. the 5/6 aspect of gearhulk establishes a 4 turn clock (flash speed means he hits the field running on your turn) and is big enough to get around many "big critter" hitters in modern like thrun, a non-maxed out goyf, etc.
So, snap is cheaper most of the time, has a wider range of targets, but is far less useful in terms of field presence. Not exactly clear when comparing them directly, as arguments can be made easily in either direction. Snap grabs more faster, gearhulk is a beast, etc. Thus, I think its worth noting that, in my opinion, the answer lies in the situation of magic's structure itself in terms of decktypes.
The point of a control deck is to say no to opponent long enough to field and complete a win condition. Meaning, the key to a control deck is finding the balance needed between control cards and win conditions so that you have just enough control stuff to stop opponent from doing what he wants but not too much so that you sit there all game waiting for a win condition to surface, giving opponent time to regroup. Reactive control decks do this with reactive effects like counterspells that buy time while waiting for a sustainable singular win card like jace the mind sculptor, for a combo to go off, or for man lands to tick away at opponent's life. Active control decks do this with effects that eliminate threats before they surface, such as discard stuff, land hate stuff, extraction stuff, creatures with "hit the field" effects, etc. They field win conditions not in 1 chunk or in simple combo but through creatures and planeswalkers that give both control and a threat of damage/a winning condition, thus getting both control and a win condition out of most of their cards. They are less efficient at either aspect however cover both necessities, which is the argument for those types of decks (which are pretty prevalent).
So, it is ultimately my contention that gearhulk is the perfect addition to a reactive control deck, and snap is the perfect addition to active control deck.
As stated above, reactive control counters stuff, fields a win condition, and completes a win condition. Gearhulk is PERFECT for this setup as he gives you control via flashbacked cards in the endgame (when you are low on material) and is a fast win condition (4 turn clock that comes in at instant speed). Because it costs so much, you won't play it early, but you don't need a win condition early in reactive control, so it fits the win condition slot but still generates control as well (retaining card advantage). Snap doesn't act as a win condition (unless its part of an elaborate combo), and thus can't be used as a win condition spot, forcing reactive control decks to use up a spot for a pure win condition.
However, active control decks can't wait for the 6 lands necessary to play gearhulk, as they hit the ground with control and win conditions simultaneously, making snapcaster the better pick for this build. It hits in the middle game, long enough to give the grave some targets, and is enough field presence to add to the already accumulating threat the active control deck tries to build.
So, to sum up, torrential gearhulk is the win condition reactive control decks dream about, and snapcaster mage is the first pick of active control decks on all fronts.
Please feel free to disagree with me, these are just my thoughts on the matter. I find myself gravitating toward reactive control decks rather than active ones, so I tend to like gearhulk better.